ALAMO: Automatic Learning of Algebraic Models for Optimization

Dr. Nicolaos Sahinidis
Department of Chemical Engineering
Carnegie Mellon University, USA

3:30pm - April 18, 2013
(ETLC) 1-001

Abstract:

We address the problem of learning algebraic models from experimental or simulation data. We present a technique for developing simple, yet accurate, models, while minimizing the number of experiments or simulations of the system under study. The methodology begins by building a low-complexity model of the system using integer optimization techniques. The model is then tested, exploited, and improved through the use of derivative-free optimization to adaptively sample new experimental or simulation points. We provide computational comparisons between ALAMO, the computational implementation of the proposed methodology, and a variety of machine learning and statistical techniques, including Latin hypercube sampling, simple least squares regression, and lasso. Finally, we present an application in the optimal design of CO2 capture systems using a detailed process simulator.

Biography:

Nick Sahinidis is John E. Swearingen Professor at Carnegie Mellon University. His research has included the development of theory, algorithms, and the BARON software for global optimization of mixed-integer nonlinear programs. Scientists and engineers have used BARON in many application areas, including the development of new Runge-Kutta methods for partial differential equations, energy policy making, modeling and design of metabolic processes, product and process design, engineering design, and automatic control. Several companies have also used BARON in the automotive, financial, and chemical process industries. Professor Sahinidis's research activities have been recognized by a National Science Foundation CAREER award in 1995, the 2004 INFORMS Computing Society Prize, the 2006 Beale-Orchard-Hays Prize from the Mathematical Programming Society, and the 2010 Computing in Chemical Engineering Award.